Abstract

Conventionally, pattern recognition problems involve both samples and features that get collected over time or that gets generated from distributed sources. The system starts to falter when the number of features reaches a certain threshold and exhibits the curse of dimensionality. Traditionally dimensionality reduction (DR) is performed to prevent the curse of dimensionality when all features are available or when the system starts to degrade in its performance. But in the current digital age systems, the enormous and continuous influx of both samples and features mandates performing DR at regular intervals to keep up with the system performance. The massive amount of feature space and sample space that gets accumulated instantaneously allows little chance to extract the knowledge effectively that can be used promptly and hence mandates performing the DR at regular intervals of time. In real time scenarios, for any domain, decisions have to be made as and when the data is made available to realize the best outcome and to mitigate the risks. The various ways in which the features flow or get generated can be different depending on the domain of the dataspace. Due to its ever changing environment, extraction of knowledge can get more challenging. To overcome this problem of big data, an incremental dimensionality reduction (IDR) approach to extract, carry forward, build and accumulate the knowledge without recalling the previous data is explored in this case study. Both Feature subsetting and Feature transformation methods are employed for the purpose of illustrating the incremental reduction of attributes. The hyperspectral image generated from an AVIRIS sensor provides a versatile environment required to demonstrate the in depth study of an IDR approach. This case study attempts to showcase a novel approach of maximizing the knowledge while minimizing the information loss through the use of IDR techniques in a multifaceted environment with hyperspectral data.

References

• https://purr.purdue.edu/publications/1947/1
• http://www.cse.msu.edu/~cse802/Feature_selection.pdf
• Plaza, J., Plaza, A. J., & Barra, C. (2009). Multi-Channel Morphological Profiles for Classification of Hyperspectral Images Using Support Vector Machines. Sensors, 9, 196-218.
• Zhang, Y., Du, B., Zhang, L., & Liu, T. (2017). Joint Sparse Representation and Multitask Learning for Hyperspectral Target Detection. IEEE Transactions on Geoscience and Remote Sensing, 55(2), 894-906.
• http://www.microimages.com/documentation/Tutorials/hyprspec.pdf
• http://lesun.weebly.com/hyperspectral-data-set.html AND https://purr.purdue.edu/publications/1947/1
• Aggarwal, C. C., & Yu, P. S. (2001, May). Outlier detection for high dimensional data. In ACM Sigmod Record (Vol. 30, No. 2, pp. 37-46). ACM.(sparsity)
• Signal Theory Methods in Multispectral Remote Sensing. David A Landgrebe. ISBN: 978-0-471-42028-6. 528 pages. January 2003
• Subramanian, S., Gat, N., Ratcliff, A., & Eismann, M. (2000). Real-time hyperspectral data compression using principal components transformation. In In Proceedings of the AVIRIS Earth Science & Applications Workshop.
• C.-I Chang, Hyperspectral Data Exploitation: Theory and Applications. New Jersey: John Wiley and Sons, 2007.
• P. Zhong, P. Zhang, and R. Wang, “Dynamic learning of SMLR for feature selection and classification of hyperspectral data,” IEEE Geoscience and Remote Sensing Letters, vol. 5, no. 2, pp. 280-284, April 2008
• Preet, P., & Batra, S. S. (2015). Feature Selection for classification of hyperspectral data by minimizing a tight bound on the VC dimension. arXiv preprint arXiv:1509.08112.
• Pal, Mahesh, and Giles M. Foody. ”Feature selection for classification of hyperspectral data by SVM.” Geoscience and Remote Sensing, IEEE Transactions on 48.5 (2010): 2297-2307.
• Agarwal, Abhishek, Tarek El-Ghazawi, Hesham El-Askary, and Jacquline Le-Moigne. ”Efficient hierarchical-PCA dimension reduction for hyperspectral imagery.” In Signal Processing and Information Technology, 2007 IEEE International Symposium on, pp. 353-356. IEEE, 2007.
• Syed Zakir Ali., P Nagabhushan., Pradeep Kumar R, Incremental datamining using Clustering Intelligent Methods of Fusing the Knowledge During Incremental Learning via Clustering in A Distributed Environment , PhD Thesis, 2010
• Meenakshi, H. N., & Nagabushan, P (2017). Target Class Guided Compression in Feature Subspace, IJCST, 4(6).
• Nagabhushan, P., & Mahadev, P. (2014). Incremental Feature Subsetting useful for Big Feature Space Problems. International Journal of Computer Applications, 97(12).
• Preeti Mahadev and P Nagabhushan. Incremental Feature Transformation for Temporal Space. International Journal of Computer Applications 145(8):28-38, July 2016
• P. Nagabhushan, An efficient method for classifying remotely sensed data (incorporating dimensionality reduction), Ph.D thesis, Universityof Mysore, 1988
• Datta, Aloke, Susmita Ghosh, and Ashish Ghosh. "Unsupervised band extraction for hyperspectral images using clustering and kernel principal component analysis." International Journal of Remote Sensing 38.3 (2017): 850-873.
• NASA JPL, AVIRIS Data Portal [online]. Available at http://aviris.jpl.nasa.gov/alt_locator/. [Accessed June 2016].
• Principal Component. Analysis, Second Edition. I.T. Joliffe. Springer, NewYork, 2002
• Liu, H., & Motoda, H. (1998). Feature transformation and subset selection. IEEE Intell Syst Their Appl, 13(2), 26-28.
• https://people.eecs.berkeley.edu/~vazirani/algorithms/chap2.pdf